Esc

↑↓ move↵ openIndex · Pagefind
Scheduling / Risk analysis: three-point estimates, risk register and correlation

Risk analysis: three-point estimates, risk register and correlation

Give activities optimistic, most likely and pessimistic durations, keep a risk register with probability, impact and mitigation, group correlated work, and run a reproducible Monte Carlo simulation that shows pre- and post-mitigation finish dates.

On this page

A Monte Carlo simulation is only as good as the ranges you give it. The Risk page (Schedule › Analyze › Risk) is where you enter those inputs: a three-point estimate for each activity, a risk register of events that might happen, and groups of activities whose durations move together. The simulation itself runs from Analysis › Risk. See Schedule analysis.

Risk inputs need Schedule admin. They’re saved with the plan, so each run is tied to an exact version of both.

The workflow

  1. Enter three-point durations. On Three-point durations, give each activity an optimistic, most likely and pessimistic value, in working days or as a percentage of remaining duration. You can paste rows straight from a spreadsheet.
  2. Fill gaps with templates. On Templates, set ranges by WBS (optionally including child nodes), activity code or trade, and apply them. Templates write percentage estimates, and leave entered or imported rows alone unless you tick Overwrite entered and imported.
  3. Record risk events. On Risk register, add the things that might happen, with their probability, impact, owner and mitigation.
  4. Correlate related work. On Correlation & key dates, put activities that tend to overrun together into a group with a coefficient, and choose up to 20 key dates to report.
  5. Save as new version. Saving creates a new version of the plan, like any other edit.
  6. Run it. Open Analysis › Risk, choose the distribution, iterations and seed, and run.

Three-point estimates

  • One estimate per activity, with optimistic ≤ most likely ≤ pessimistic.
  • Working days (0 to 100,000) or percent of remaining duration (−100% to 1,000%). In a pasted row, a trailing % means percent, and all three values must use the same unit. Rows can be separated by tabs, commas or semicolons.
  • Each estimate records its basis: entered, template or imported.
  • On a started activity, the estimate covers the remaining duration. Estimates on completed activities, milestones, level-of-effort and WBS summaries are ignored.

Activities without an estimate either use the engine’s default spread (−10% / +45%, which has no stated source) or are held at their planned duration. You choose which. Every result states where its estimates came from: real, mixed, default spread or fixed.

The distribution is chosen per run, not per activity: PERT (beta), the default, or Triangular.

Risk register

Each event has:

  • A title, description, owner and category.
  • A status: Open (simulated) or Closed (not simulated).
  • A mode: a duration impact on each activity it affects, or a fragnet ahead of each activity, which inserts the delay before it.
  • Affected activities, up to 500.
  • Pre-mitigation probability (0 to 1) and minimum, most likely and maximum impact, in days or percent.
  • Mitigation text and optional post-mitigation probability and impact.

Results rank events by their contribution to the finish.

Correlation

Activities in a correlation group share a coefficient from 0 to 0.99, so their durations tend to run long or short together. Each activity belongs to at most one group. The model uses a one-factor Gaussian copula: correlation is equal within a group, and there’s no matrix between groups. The result reports the correlation actually achieved in each group, so you can see whether the target held.

Weather

The Weather tab adds a weather allowance to the simulation, so lost days fall on the work the weather actually stops.

  • Weather calendars: for each region, the expected working days lost to weather in each month, with an optional minimum and maximum (by default 0 and twice the expected days). A blank month allows nothing. Up to 50 regions; choose the one this job uses.
  • Import them as CSV: region, period, expected days[, min days, max days], where the period is a month (1–12 or Jan…Dec) or a season (winter, spring, summer, autumn). A later month row refines an earlier season row.
  • Weather-sensitive activities, listed by ID, or picked by an activity code value such as Exposure=External. Other work isn’t touched.
  • How weather is modelled:
    • Calendar exceptions adds the region’s expected days as non-working days, spread evenly through each month, on a copy of each sensitive activity’s calendar.
    • Probabilistic draws each month’s lost days in every iteration from its minimum, expected and maximum. One draw per month is shared by all sensitive activities, and each gets a share in proportion to the working days it has in that month.

Either way, the result’s Weather impact shows the finish percentiles with and without weather on the same random numbers, the difference in calendar days, and the sensitive activities ranked by expected lost working days.

What a run shows

  • Finish: the deterministic date, mean and standard deviation, and P10, P20, P50, P80, P90 and P95, with a finish date S-curve, a percentile table and a histogram.
  • Drivers: a sensitivity tornado of the top 20 activities, and the criticality index of each activity.
  • Key dates: the same percentiles for each key date you chose.
  • Pre- vs post-mitigation: when any event has post-mitigation values, a second scenario runs with the same random numbers, and the page shows how many days mitigation buys at each percentile. This doubles the run time.
  • Per activity: up to 200 activities with criticality index, sensitivity, mean duration and the estimate used.
  • Risk provenance: the seed (and whether you chose it), iterations, distribution, a digest of the estimates and of the model, and the engine revision.

Reproducible by design

Enter a seed and the same inputs always give the same answer. Leave the seed blank and one is generated and recorded. The result tells you to rerun with that seed to reproduce it. Events and correlation draw from separate random streams, so adding an event doesn’t reshuffle the duration samples.

Iterations run from 100 to 20,000; the page defaults to 2,000. Large runs (activities × iterations over 500,000) run as a background job with progress and cancel.

Exchanging estimates with P6

The P6 interchange tab writes the three-point estimates as three numeric activity UDFs: Risk Optimistic Duration (d), Risk Most Likely Duration (d) and Risk Pessimistic Duration (d). These travel in XER and P6 XML. Turning imported UDF values into estimates is an explicit step on that tab.

Limits

  • An event’s contribution is a conditional-mean estimate, not a re-run with the event removed. When events overlap, their shares are approximate.
  • Events are independent of each other and of duration uncertainty. Each occurrence uses one impact draw.
  • Lags, constraints and calendars are never sampled, apart from the lost days of the weather allowance.
  • Weather calendars and weather-sensitive activities aren’t exported to P6 or MSPDI.
  • The risk register, correlation groups and templates aren’t exported to P6 or MSPDI. Only the three estimate UDFs are.
  • Model size: 10,000 estimates, 200 events, 100 correlation groups (1,000 activities each), 200 templates and 20 key dates.
  • There’s no REST endpoint for editing risk inputs. Runs are available through POST /schedule/calculations with operation: "risk".
  • Mistake: Publishing a P80 from the default spread

    In the app: Check the estimates source banner. A result built on the unsourced default spread is a sensitivity test, not a forecast.

  • Mistake: Adding the same delay as a risk event and inside the pessimistic duration

    In the app: Keep known variability in three-point durations and discrete events in the register, so the delay isn’t counted twice.

  • Mistake: Comparing two runs with different seeds

    In the app: Fix the seed when you compare mitigation options, or rely on the built-in pre- and post-mitigation comparison.

Keep exploring

Related across os.construction

Run it on your own project.

Start a 14-day free trial. Want a say in what ships next? Join the founding program.

Your role (optional)

14-day free trial. Pricing for paid plans isn't set yet. Talk to us. Already have an account? Sign in →

Ask the OS

Reading this page

Ask anything about this page. Answers use what you're reading as context.

Suggested

    AI · may be imperfect · sample data where shown

    Open file